Can AI identify bias in media

Nowhere News is a recent news service attempting to use artificial intelligence to identify bias in news. The service accesses a large number of news services and identifies popular stories of the day. The service than uses the content from multiple services to create a neutral account and also generates a right-wing and a left-wing version (or positive and negative view if more appropriate). The developers have weighted the sites they access by reputation for accuracy and do review and edit the generated summaries before publishing. 

I read multiple descriptions of Nowhere News to try to understand more about what the AI is doing [ TechCrunch, Singularity Hub, Vice]. Services that summarize web content have been around for several years. One approach might involve summarizing multiple accounts from news services known to have right, left, and neutral biases. This does not seem to be what is happening here. As I understand AI, you input many signals and identify a characteristic of the source (in this case, right, neutral, and left) and the system learns how to use the signals to more and more accurately predict the designated category. I am not certain this appropriately summarizes what has been done here.

However generated, the three versions of a story generalized from multiple sources would seem to have some educational value. Just reading the three summaries offers readers insight into what bias means. [see Common Sense Media comment on educational potential].

At present the Knowhere News site is free and I would assume would eventually need to find some way to monetize. If you enroll as a user, you receive emails with story summarizes. You can also just visit the site and read the stories summarized for a given day.

This service is interesting and useful. I doubt it could be regarded as a solution to the fake news crisis.  First, I don’t see brief summaries the equivalent of reading long-form news stories. Knowhere does provide links to multiple sources, but often not to sources I would read. I am not clear how major news sources with some version of a pay to read model regard content being used in this way. Second and as acknowledged by Knowhere, there is a difference between bias (spin) and quality. See my previous post.